Regarding a network system, load balancing is one of important issues. For example, regarding a nonstructural network such as an enterprise network, a technique described in Non-Patent Literature 1 achieves comprehensive load balancing by considering a network congestion state and a server load state.
Patent Literature 1 (Japanese Patent Publication JP-2005-64882) discloses a higher-order layer processing system. In the higher-order layer processing system, a plurality of higher-order layer processing devices are arranged in parallel in order to secure scalability of the system. In order to equally use resources of the higher-order layer processing devices, a packet distribution device is provided at a former stage of the higher-order layer processing devices. The packet distribution device distributes packets to the plurality of higher-order layer processing devices on the basis of information of L3 or lower of a received packet. More specifically, the packet distribution device incorporates a hash into the received packet and sends the packet to the higher-order layer processing device. The higher-order layer processing device receives the packet and, if there is a possibility that the device itself becomes the congestion state, informs the packet distribution device accordingly. In this case, the packet distribution device selects another higher-order layer processing device. As a result, load centralization to a particular higher-order layer processing device can be prevented.